Mini Shell
""" Convenience Classes building on the base PMAPI extension module """
#
# Copyright (C) 2013-2016,2019 Red Hat
# Copyright (C) 2009-2012 Michael T. Werner
#
# This file is part of the "pcp" module, the python interfaces for the
# Performance Co-Pilot toolkit.
#
# This program is free software; you can redistribute it and/or modify it
# under the terms of the GNU General Public License as published by the
# Free Software Foundation; either version 2 of the License, or (at your
# option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
# for more details.
#
# pylint: disable=line-too-long,protected-access,dangerous-default-value
# pylint: disable=too-many-lines,too-many-arguments,too-many-nested-blocks
# pylint: disable=consider-using-dict-items
#
import sys
from ctypes import c_int, c_uint, c_char_p, cast, POINTER
from pcp.pmapi import pmContext, pmValue, pmDesc, pmErr, timeval
from cpmapi import (PM_CONTEXT_HOST, PM_CONTEXT_ARCHIVE, PM_INDOM_NULL,
PM_IN_NULL, PM_ID_NULL, PM_SEM_COUNTER, PM_ERR_EOL,
PM_TYPE_DOUBLE)
class MetricCore(object):
"""
Core metric information that can be queried from the PMAPI
PMAPI metrics are unique by name, and MetricCores should be also
rarely, some PMAPI metrics with different names might have identical PMIDs
PMAPI metrics are unique by (name) and by (name,pmid) - _usually_ by (pmid)
too. Note that names here (and only here) are stored as byte strings for
direct PMAPI access. All dictionaries/caching strategies built using the
core structure use native strings (i.e., not byte strings in python3).
"""
def __init__(self, ctx, name, pmid):
self.ctx = ctx
if not isinstance(name, bytes):
name = name.encode('utf-8')
self.name = name
self.pmid = pmid
self.desc = None
self.text = None
self.help = None
class Metric(object):
"""
Additional metric information, such as conversion factors and values
several instances of Metric may share a MetricCore instance
"""
##
# constructor
def __init__(self, core):
self._core = core # MetricCore
self._vset = None # pmValueSet member
self._values = None
self._prevvset = None
self._prevValues = None
self._convType = core.desc.contents.type
self._convUnits = None
self._errorStatus = None
self._netValues = None # (instance, name, value)
self._netPrevValues = None # (instance, name, value)
self._netConvertedValues = None # (instance, name, value)
##
# core property read methods
def _R_ctx(self):
return self._core.ctx
def _R_name(self):
return self._core.name.decode()
def _R_pmid(self):
return self._core.pmid
def _R_desc(self):
return self._core.desc
def _R_text(self):
return self._core.text
def _R_help(self):
return self._core.help
def get_vlist(self, vset, vlist_idx):
""" Return the vlist[vlist_idx] of vset """
listptr = cast(vset.contents.vlist, POINTER(pmValue))
return listptr[vlist_idx]
def get_inst(self, vset, vlist_idx):
""" Return the inst for vlist[vlist_idx] of vset """
return self.get_vlist(vset, vlist_idx).inst
def computeValues(self, inValues):
""" Extract the value for a singleton or list of instances
as a triple (inst, name, val)
"""
vset = inValues
ctx = self.ctx
instD = ctx.mcGetInstD(self.desc.contents.indom)
valL = []
for i in range(vset.numval):
instval = self.get_vlist(vset, i)
try:
name = instD[instval.inst]
except KeyError:
name = ''
outAtom = self.ctx.pmExtractValue(vset.valfmt, instval,
self.desc.type, self._convType)
if self._convUnits:
desc = (POINTER(pmDesc) * 1)()
desc[0] = self.desc
outAtom = self.ctx.pmConvScale(self._convType, outAtom, desc,
0, self._convUnits)
value = outAtom.dref(self._convType)
valL.append((instval, name, value))
return valL
def _find_previous_instval(self, index, inst, pvset):
""" Find a metric instance in the previous resultset """
if index <= pvset.numval:
pinstval = self.get_vlist(pvset, index)
if inst == pinstval.inst:
return pinstval
for pi in range(pvset.numval):
pinstval = self.get_vlist(pvset, pi)
if inst == pinstval.inst:
return pinstval
return None
def convertValues(self, values, prevValues, delta):
""" Extract the value for a singleton or list of instances as a
triple (inst, name, val) for COUNTER metrics with the value
delta calculation applied (for rate conversion).
"""
if self.desc.sem != PM_SEM_COUNTER:
return self.computeValues(values)
if prevValues is None:
return None
pvset = prevValues
vset = values
ctx = self.ctx
instD = ctx.mcGetInstD(self.desc.contents.indom)
valL = []
for i in range(vset.numval):
instval = self.get_vlist(vset, i)
pinstval = self._find_previous_instval(i, instval.inst, pvset)
if pinstval is None:
continue
try:
name = instD[instval.inst]
except KeyError:
name = ''
outAtom = self.ctx.pmExtractValue(vset.valfmt, instval,
self.desc.type, PM_TYPE_DOUBLE)
poutAtom = self.ctx.pmExtractValue(pvset.valfmt, pinstval,
self.desc.type, PM_TYPE_DOUBLE)
if self._convUnits:
desc = (POINTER(pmDesc) * 1)()
desc[0] = self.desc
outAtom = self.ctx.pmConvScale(PM_TYPE_DOUBLE, outAtom, desc,
0, self._convUnits)
poutAtom = self.ctx.pmConvScale(PM_TYPE_DOUBLE, poutAtom, desc,
0, self._convUnits)
value = outAtom.dref(PM_TYPE_DOUBLE)
pvalue = poutAtom.dref(PM_TYPE_DOUBLE)
if value >= pvalue:
valL.append((instval, name, (value - pvalue) / delta))
return valL
def _R_values(self):
return self._values
def _R_prevValues(self):
return self._prevValues
def _R_convType(self):
return self._convType
def _R_convUnits(self):
return self._convUnits
def _R_errorStatus(self):
return self._errorStatus
def _R_netConvValues(self):
return self._netConvValues
def _R_netPrevValues(self):
if not self._prevvset:
return None
self._netPrevValues = self.computeValues(self._prevvset)
return self._netPrevValues
def _R_netValues(self):
if not self._vset:
return None
self._netValues = self.computeValues(self._vset)
return self._netValues
def _W_values(self, values):
self._prev = self._values
self._values = values
self._netPrev = self._netValue
self._netValue = None
def _W_convType(self, value):
self._convType = value
def _W_convUnits(self, value):
self._convUnits = value
# interface to properties in MetricCore
ctx = property(_R_ctx, None, None, None)
name = property(_R_name, None, None, None)
pmid = property(_R_pmid, None, None, None)
desc = property(_R_desc, None, None, None)
text = property(_R_text, None, None, None)
help = property(_R_help, None, None, None)
# properties specific to this instance
values = property(_R_values, _W_values, None, None)
prevValues = property(_R_prevValues, None, None, None)
convType = property(_R_convType, _W_convType, None, None)
convUnits = property(_R_convUnits, _W_convUnits, None, None)
errorStatus = property(_R_errorStatus, None, None, None)
netValues = property(_R_netValues, None, None, None)
netPrevValues = property(_R_netPrevValues, None, None, None)
netConvValues = property(_R_netConvValues, None, None, None)
def metricPrint(self):
indomstr = self.ctx.pmInDomStr(self.desc.indom)
print(" ", "indom:", indomstr)
_ = self.ctx.mcGetInstD(self.desc.indom)
for _, name, val in self.netValues:
print(" ", name, val)
def metricConvert(self, delta):
convertedList = self.convertValues(self._vset, self._prevvset, delta)
self._netConvValues = convertedList
return self._netConvValues
class MetricCache(pmContext):
"""
A cache of MetricCores is kept to reduce calls into the PMAPI library
this also slightly reduces the memory footprint of Metric instances
that share a common MetricCore
a cache of instance domain information is also kept, which further
reduces calls into the PMAPI and reduces the memory footprint of
Metric objects that share a common instance domain
"""
##
# overloads
def __init__(self, typed=PM_CONTEXT_HOST, target="local:"):
pmContext.__init__(self, typed, target)
self._mcIndomD = {}
self._mcByNameD = {}
self._mcByPmidD = {}
self._mcCounter = 0 # number of counters in this cache
self._mcMetrics = 0 # number of metrics in this cache
##
# methods
def mcGetInstD(self, indom):
""" Query the instance : instance_list dictionary """
return self._mcIndomD[indom]
def _mcAdd(self, core):
""" Update the dictionary """
indom = core.desc.contents.indom
if indom not in self._mcIndomD:
if c_int(indom).value == c_int(PM_INDOM_NULL).value:
instmap = {PM_IN_NULL: b'PM_IN_NULL'}
else:
if self._type == PM_CONTEXT_ARCHIVE:
instL, nameL = self.pmGetInDomArchive(core.desc)
else:
instL, nameL = self.pmGetInDom(core.desc)
if instL is not None and nameL is not None:
instmap = dict(zip(instL, nameL))
else:
instmap = {}
self._mcIndomD.update({indom: instmap})
if core.desc.contents.sem == PM_SEM_COUNTER:
self._mcCounter += 1
self._mcMetrics += 1
self._mcByNameD.update({core.name.decode(): core})
self._mcByPmidD.update({core.pmid: core})
def mcGetCoresByName(self, nameL):
""" Update the core (metric id, description,...) list """
coreL = []
missD = None
errL = None
# lookup names in cache
for index, name in enumerate(nameL):
if isinstance(name, bytes):
name = name.decode()
# lookup metric core in cache
core = self._mcByNameD.get(name)
if not core:
# cache miss
if not missD:
missD = {}
missD.update({name: index})
coreL.append(core)
# some cache lookups missed, fetch pmids and build missing MetricCores
if missD:
idL, errL = self.mcFetchPmids(missD.keys())
for name, pmid in idL:
if pmid == PM_ID_NULL:
# fetch failed for the given metric name
if not errL:
errL = []
errL.append(name)
else:
# create core pmDesc
newcore = self._mcCreateCore(name, pmid)
# update core ref in return list
coreL[missD[name]] = newcore
return coreL, errL
def _mcCreateCore(self, name, pmid):
""" Update the core description """
newcore = MetricCore(self, name, pmid)
try:
newcore.desc = self.pmLookupDesc(pmid)
except pmErr as error:
fail = "%s: pmLookupDesc: %s" % (error.progname(), error.message())
sys.stderr.write(fail)
raise SystemExit(1)
# insert core into cache
self._mcAdd(newcore)
return newcore
def mcFetchPmids(self, nameL):
""" Update the core metric ids. note: some names have identical pmids """
errL = None
nameA = (c_char_p * len(nameL))()
for index, name in enumerate(nameL):
if not isinstance(name, bytes):
name = name.encode('utf-8')
nameA[index] = c_char_p(name)
try:
pmidArray = self.pmLookupName(nameA)
if len(pmidArray) < len(nameA):
missing = "%d of %d metric names" % (len(pmidArray), len(nameA))
sys.stderr.write("Cannot resolve %s" % missing)
raise SystemExit(1)
except pmErr as error:
fail = "%s: pmLookupName: %s" % (error.progname(), error.message())
sys.stderr.write(fail)
raise SystemExit(1)
return zip(nameL, pmidArray), errL
class MetricGroup(dict):
"""
Manages a group of metrics for fetching the values of
a MetricGroup is a dictionary of Metric objects, for which data can
be fetched from a target system using a single call to pmFetch
the Metric objects are indexed by the metric name
pmFetch fetches data for a list of pmIDs, so there is also a shadow
dictionary keyed by pmID, along with a shadow list of pmIDs
"""
##
# property read methods
def _R_contextCache(self):
return self._ctx
def _R_nonCounters(self):
return self._ctx._mcCounter != self._ctx._mcMetrics
def _R_pmidArray(self):
return self._pmidArray
def _R_timestamp(self):
return self._result.contents.timestamp
def _R_result(self):
return self._result
def _R_prevTimestamp(self):
return self._prev.contents.timestamp
def _R_prev(self):
return self._prev
##
# property write methods
def _W_result(self, pmresult):
self._prev = self._result
self._result = pmresult
##
# property definitions
contextCache = property(_R_contextCache, None, None, None)
nonCounters = property(_R_nonCounters, None, None, None)
pmidArray = property(_R_pmidArray, None, None, None)
result = property(_R_result, _W_result, None, None)
timestamp = property(_R_timestamp, None, None, None)
prev = property(_R_prev, None, None, None)
prevTimestamp = property(_R_prevTimestamp, None, None, None)
##
# overloads
def __init__(self, contextCache, inL=[]):
dict.__init__(self)
self._ctx = contextCache
self._pmidArray = None
self._result = None
self._prev = None
self._altD = {}
self.mgAdd(inL)
def __setitem__(self, attr, value):
if attr in self:
raise KeyError("metric group with that key already exists")
dict.__setitem__(self, attr, MetricGroup(self, inL=value))
##
# methods
def mgAdd(self, nameL):
""" Create the list of Metric(s) """
coreL, _ = self._ctx.mcGetCoresByName(nameL)
for core in coreL:
metric = Metric(core)
self.update({metric.name: metric})
self._altD.update({metric.pmid: metric})
n = len(self)
self._pmidArray = (c_uint * n)()
for x, key in enumerate(self.keys()):
self._pmidArray[x] = c_uint(self[key].pmid)
def mgFetch(self):
"""
Fetch the list of Metric values. Save the old value.
Return the result timestamp.
"""
try:
self.result = self._ctx.pmFetch(self._pmidArray)
# update the result entries in each metric
result = self.result.contents
for i in range(self.result.contents.numpmid):
pmid = self.result.contents.get_pmid(i)
vset = self.result.contents.get_vset(i)
self._altD[pmid]._prevvset = self._altD[pmid]._vset
self._altD[pmid]._vset = vset
except pmErr as error:
if error.args[0] == PM_ERR_EOL:
raise SystemExit(0)
fail = "%s: pmFetch: %s" % (error.progname(), error.message())
sys.stderr.write(fail)
raise SystemExit(1)
return result.timestamp # timeval
def mgDelta(self):
"""
Sample delta - used for rate conversion calculations, which
requires timestamps from successive samples.
"""
if self._prev is not None:
prevTimestamp = float(self.prevTimestamp)
else:
prevTimestamp = 0.0
return float(self.timestamp) - prevTimestamp
class MetricGroupPrinter(object):
"""
Handles reporting of MetricGroups within a GroupManager.
This object is called upon at the end of each fetch when
new values are available. It is also responsible for
producing any initial (or on-going) header information
that the tool may wish to report.
"""
def report(self, manager):
""" Base implementation, all tools should override """
for group_name in manager.keys():
group = manager[group_name]
for metric_name in group.keys():
group[metric_name].metricPrint()
def convert(self, manager):
""" Do conversion for all metrics across all groups """
for group_name in manager.keys():
group = manager[group_name]
delta = group.mgDelta()
for metric_name in group.keys():
group[metric_name].metricConvert(delta)
class MetricGroupManager(dict, MetricCache):
"""
Manages a dictionary of MetricGroups which can be pmFetch'ed
inherits from MetricCache, which inherits from pmContext
"""
##
# property access methods
def _R_options(self): # command line option object
return self._options
def _W_options(self, options):
self._options = options
def _R_default_delta(self): # default interval unless command line set
return self._default_delta
def _W_default_delta(self, delta):
self._default_delta = delta
def _R_default_pause(self): # default reporting delay (archives only)
return self._default_pause
def _W_default_pause(self, pause):
self._default_pause = pause
def _W_printer(self, printer): # helper class for reporting
self._printer = printer
def _R_counter(self): # fetch iteration count, useful for printer
return self._counter
##
# property definitions
options = property(_R_options, _W_options, None, None)
default_delta = property(_R_default_delta, _W_default_delta, None, None)
default_pause = property(_R_default_pause, _W_default_pause, None, None)
printer = property(None, _W_printer, None, None)
counter = property(_R_counter, None, None, None)
##
# overloads
def __init__(self, typed=PM_CONTEXT_HOST, target="local:"):
dict.__init__(self)
MetricCache.__init__(self, typed, target)
self._options = None
self._default_delta = timeval(1, 0)
self._default_pause = None
self._printer = None
self._counter = 0
def __setitem__(self, attr, value):
if attr in self:
raise KeyError("metric group with that key already exists")
dict.__setitem__(self, attr, MetricGroup(self, inL=value))
@classmethod
def builder(cls, options, argv):
""" Helper interface, simple PCP monitor argument parsing. """
manager = cls.fromOptions(options, argv)
manager._default_delta = timeval(options.delta, 0)
manager._options = options
return manager
##
# methods
def _tv2float(self, tv):
""" convert timeval to epoch seconds as a float """
if tv is None:
return 0.0
return float(tv.tv_sec) + float(tv.tv_usec) / 1e6
def _computeSamples(self):
""" Return the number of samples we are to take and the
finish time, or 0,0 if --finish is not specified.
This is based on command line options --samples but also
must consider --start, --finish and --interval. If none
of these were presented, a zero return means "infinite".
Also consider whether the utility needs rate-conversion,
automatically increasing the sample count to accommodate
when counters metrics are present.
"""
if self._options is None:
return 0, None # loop until interrupted or PM_ERR_EOL
extra = 1 # extra sample needed if rate converting
for group in self.keys():
if self[group].nonCounters:
extra = 0
samples = self._options.pmGetOptionSamples()
if samples is not None:
return samples + extra, None
if self._options.pmGetOptionFinishOptarg() is None:
return 0, None # loop until interrupted or PM_ERR_EOL
origin = self._options.pmGetOptionOrigin()
finish = self._options.pmGetOptionFinish()
delta = self._options.pmGetOptionInterval()
if delta is None:
delta = self._default_delta
period = self._tv2float(delta)
window = (self._tv2float(finish) - self._tv2float(origin)) / period
# return samples rounded to positive number and the finish time as a float
return int(window + 0.5) + extra, finish
def _computePauseTime(self):
""" Figure out how long to sleep between samples.
This needs to take into account whether we were explicitly
asked for a delay (independent of context type, --pause),
whether this is an archive or live context, and the sampling
--interval (including the default value, if none requested).
"""
if self._default_pause is not None:
return self._default_pause
if self.type == PM_CONTEXT_ARCHIVE:
self._default_pause = timeval(0, 0)
elif self._options is not None:
pause = self._options.pmGetOptionInterval()
if pause is not None:
self._default_pause = pause
else:
self._default_pause = self._default_delta
else:
self._default_pause = self._default_delta
return self._default_pause
def checkMissingMetrics(self, nameL):
"""
Return a list of metrics that are missing from the default context.
This is usually only applicable when replaying archives.
Return None if all found, else a list of missing metric names.
"""
missing = []
nameA = (c_char_p * len(nameL))()
for i, n in enumerate(nameL):
if not isinstance(n, bytes):
n = n.encode('utf-8')
nameA[i] = c_char_p(n)
try:
# fast path: check all in one call
pmContext.pmLookupName(self, nameA)
except pmErr:
# failure: check all names individually
for i, n in enumerate(nameA):
try:
pmContext.pmLookupName(self, (n))
except pmErr:
missing.append(nameL[i])
if not missing:
return None
return missing
def fetch(self):
""" Perform fetch operation on all of the groups. """
fetchtime = None
rmax = 0.0
for group in self.keys():
stamp = self[group].mgFetch()
if fetchtime is None or self._tv2float(stamp) > rmax:
fetchtime = stamp
rmax = self._tv2float(stamp)
return fetchtime
def run(self):
""" Using options specification, loop fetching and reporting,
pausing for the requested time interval between updates.
Transparently handles archive/live mode differences.
Note that this can be different to the sampling interval
in archive mode, but is usually the same as the sampling
interval in live mode.
"""
samples, finish = self._computeSamples()
# print("DEBUG samples=" + str(samples) + " finish=" + str(self._tv2float(finish)))
timer = self._computePauseTime()
try:
curtime = self.fetch()
while True:
if self._counter >= samples > 0:
break
if finish is not None and self._tv2float(curtime) >= self._tv2float(finish):
break
self._printer.report(self)
timer.sleep()
curtime = self.fetch()
self._counter += 1
except SystemExit as code:
return code
except KeyboardInterrupt:
pass
return 0
Zerion Mini Shell 1.0